one-line api
call any model with a single request — no gpu setup, no docker, no yaml.
20,000+ models live
run any model.
ship anything.
blaze runs 20,000+ open-source ai models on serverless gpus. one api key, one line of code, zero infrastructure.
$ pip install blaze
$ blaze run flux-schnell --input photo.png
done · output ready in 2.1s on nvidia h100trending this week
features
everything you need to ship ai
fine-tune in minutes
upload a dataset, get a custom model before your coffee compiles.
deploy anywhere
turn any run into an autoscaling production endpoint.
how it works
from prompt to production in three steps
pick a model
browse the gallery or search 20,000+ community models by task, license or hardware.
run it
every run boots on serverless gpus in seconds.
blaze run flux-schnell --input prompt.txtship it
push any run to a scalable endpoint with one flag: --deploy
testimonials
the community says it best
“blaze replaced about a quarter of our infra work with one api key — we deleted three terraform files the same afternoon.”
“fine-tuning went from ssh and prayer to a form and a coffee — our custom caption model trained in eleven minutes.”
“our render pipeline went from 40 seconds to 2 and costs dropped 80%. the gallery is my startup menu.”
pricing
pricing that scales with your curiosity
$0/mo
for weekend experiments
- 10,000 free runs
- community models
- browser playground
- api access
$19/mo
for shipping side projects
- unlimited runs
- fine-tuning
- private models
- priority gpu queue
- production endpoints
$99/mo
for serious deployments
- everything in pro
- sso + audit logs
- usage analytics
- dedicated support
faq
questions, answered
blaze is a playground and api for running open-source ai models. pick a model from the gallery (or bring your own weights), send one request, and get the output back — we handle the gpus.
no. every run executes on serverless gpus we manage. you can also pin a specific gpu type if a model needs one.
yes. outputs are yours, subject to each model's underlying license — we surface licenses on every model page so you can check before you ship.
you get 10,000 free runs each month on the hobby plan. beyond that, runs bill per second of compute at each model's listed rate — no seats, no minimums.
yes. push your weights or a fine-tune, add a readme and a license, and it appears in the community gallery with a live badge within minutes.